It is the FP8 version of it. It functions pretty much identically in terms of how you prompt it and what parameters you use but requires less storage space and less VRAM then the original version. I mainly uploading this because I want images gened with the FP8 version to be associated with this model. If you upload an image generated with this model, the on-site metadata viewer sees a foreign hash (This has a different hash than regular schnell) so they never get put under the main flux page. So hopefully images generated with this model will show up here.

If you're referring to the FP8 version then I can upload that later on too.

You don't have to wait on me though You can just download it here:

https://huggingface.co/AiAF/flux1-dev-fp8.safetensors

Keep in mind that there are several versions of these. Some are small as two gigabytes. Some are larger. There are no official FP8 versions per se, just like there's no official version of a character Lora. You'll have to pick the one that fits your needs. Larger models tend to have higher accuracy while requiring more resources and time to generate. Smaller ones take up less resources and memory and generate faster but have lower quality (according to the people who upload them)

huggingface.co/city96/FLUX.1-schnell-gguf/tree/main

Also I should mention, I originally found them here:

https://huggingface.co/Comfy-Org/flux1-schnell/blob/main/flux1-schnell-fp8.safetensors